CN101615292B - Accurate positioning method for human eye on the basis of gray gradation information - Google Patents
Accurate positioning method for human eye on the basis of gray gradation information Download PDFInfo
- Publication number
- CN101615292B CN101615292B CN2009100947640A CN200910094764A CN101615292B CN 101615292 B CN101615292 B CN 101615292B CN 2009100947640 A CN2009100947640 A CN 2009100947640A CN 200910094764 A CN200910094764 A CN 200910094764A CN 101615292 B CN101615292 B CN 101615292B
- Authority
- CN
- China
- Prior art keywords
- center
- image
- gray
- eye
- human eye
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention relates to an accurate positioning method for human eye on the basis of gray gradation information, belonging to the technical field of information. The method comprises: (1) human face detection: the area of a human face is positioned from an input image; (2) gray gradation column diagram analysis: gray gradation column diagram analysis is carried out on the human face image to determine the gray gradation range of the complexion area of the human face; (3) image strengthening: gray gradation adjustment is carried out on the image to enable eye characteristics to be more obvious; (4) Gabor wavelet filtration: Gabor wavelet filtration is carried out on the strengthened image, a real part image filtered by Gabor is synthesized to obtain a reference image; (5) cluster analysis: the synthesized reference image is analyzed by using the K-Means cluster method to obtain a binarization human eye candidate window, and the white area in each human eye candidate window after binarization is analyzed to determine the rough position of the human eye, so that the human eye window is obtained; (6) neighbourhood operation: the human eye window is scanned by two neighbourhood operators to determine the canter position of the pupilla. The invention has the advantages of simple detection method and accurate eye positioning.
Description
Technical field:
The present invention relates to a kind of human eye accurate positioning method, belong to areas of information technology based on half-tone information.
Background technology:
The structure of Automatic face recognition system has become the popular research field in a pattern-recognition and the computer vision.In general, the Automatic face recognition system mainly comprises two parts, be that people's face detects and recognition of face, at present, two parts have all had some satisfactory method, though detecting, people's face obtained reasonable effect, but because people's face exists all size and the different anglecs of rotation, make us before carrying out recognition of face, must carry out normalized to detected facial image, in the various facial characteristics of people's face, eyes are to carry out the normalized preferred features of people's face, in recognition of face, the accuracy of human eye location is to the effect important influence of recognition of face, and existing eye detection method can not position eyes accurately, and is particularly lower in facial image resolution, often the error of location is bigger under the bigger situation of noise jamming, thereby has influenced the performance of follow-up face identification system.
Summary of the invention:
The objective of the invention is to overcome the deficiency of prior art, and provide a kind of human eye pinpoint method based on half-tone information.
Technical scheme of the present invention is:
Based on the human eye accurate positioning method of half-tone information, comprise the detection of people's face, grey level histogram analysis, figure image intensifying, Gabor wavelet filtering, cluster analysis and neighborhood operation step; Specific as follows:
(1) people's face detects: with the Adaboost algorithm digital picture of input is carried out people's face detecting operation, obtain the position and the size of people's face;
(2) grey level histogram analysis: detected facial image unification is zoomed to 130 * 150 pixel sizes carry out the gray processing operation, the rectangular area that a size selecting the facial image middle position is 31 * 13 pixels is as the object of colour of skin sample analysis, if I is (i, j) be (i in the facial image, j) gray-scale value of locating, m is the average gray of rectangular area, if m<180, then in the grey level histogram of this rectangular area, find [m-10, m+10] scope in the maximum gray-scale value of occurrence number as the gray-scale value si of the colour of skin, facial image is carried out the grey level histogram analysis, determine the tonal range of skin area of skin color in people's face;
(3) figure image intensifying: the gray-scale value si of definite colour of skin, use following formula that image is strengthened operation, new gray level image (i, the gray-scale value of j) locating be designated as I ' (i, j).
c
1=2c
2,c
2=2.35
If m 〉=180 then do not strengthen operation to image, promptly I ' (i, j)=I (i, j);
(4) Gabor wavelet filtering: for the facial image after strengthening, use the Gabor small echo that image is carried out filtering operation, use formula to be
In the formula, z=(x, y) the spatial domain coordinate of remarked pixel point, || || the computing of expression norm,
k
v=k
Max/ f
v,
V ∈ 0,1 ..., 4} and μ ∈ { 0,1, ..., 7} represents the direction and the yardstick of Gabor wave filter respectively, in 40 real part images that produce after Gabor filtering, selects μ=0, v={2,3,4}, promptly on the horizontal direction three of the yardstick maximum real part images according to formula:
G(x,y)=q
1f(x,y)*ψ
2,0+q
2f(x,y)*ψ
3,0+q
3f(x,y)*ψ
4,0
The reference picture that synthesizes to the end;
(5) cluster analysis: adopt the method for K-Means cluster that synthetic reference picture is analyzed, obtain the human eye candidate window, binaryzation human eye candidate window is analyzed the white portion in each the human eye candidate window after the binaryzation, obtains the human eye window;
(6) neighborhood operation: use two neighborhood operators that the human eye window is scanned, determine pupil center location.
The present invention compared with prior art has that detection method is simple, the accurate advantage of eye location.
Description of drawings:
Fig. 1 is a process flow diagram synoptic diagram of the present invention.
Fig. 2 carries out grey level histogram analytic process synoptic diagram for facial image.
Fig. 3 is a figure image intensifying process synoptic diagram.
Fig. 4 is the reference picture behind the Gabor wavelet filtering.
Fig. 5 analyzes the human eye window that obtains for the method that adopts the K-Means cluster to synthetic reference picture.
Fig. 6 is for to scan the human eye window with two neighborhood operators, the pupil center location of determining.
Embodiment:
At first use the Adaboost algorithm to carry out people's face detecting operation to the digital picture of input, obtain the position and the size of people's face, the unification of detected people's face is zoomed to 130 * 150 pixel sizes, facial image is carried out the gray processing operation, the rectangular area that a size selecting the facial image middle position is 31 * 13 pixels is as the object of colour of skin sample analysis, if I is (i, j) be (i in the facial image, j) gray-scale value of locating, m is the average gray of rectangular area, if m<180, then in the grey level histogram of this rectangular area, find [m-10, m+10] scope in the maximum gray-scale value of occurrence number as the gray-scale value si of the colour of skin, use following formula that image is strengthened operation, new gray level image is at (i, j) gray-scale value of locating be designated as I ' (i, j).
c
1=2c
2,c
2=2.35
If m 〉=180 then do not strengthen operation to image, promptly I ' (i, j)=(i, j), detailed process is seen Fig. 2 to I.For the facial image after strengthening, use the Gabor small echo that image is carried out filtering operation, use formula to be
In the formula, z=(x, y) the spatial domain coordinate of remarked pixel point, || || the computing of expression norm,
k
v=k
Max/ f
v,
V ∈ 0,1 ..., 4} and μ ∈ 0,1 ..., 7} represents the direction and the yardstick of Gabor wave filter respectively.In 40 real part images that after Gabor filtering, produce, select μ=0, v={2,3,4}, promptly three of the yardstick maximum real part images are seen Fig. 3 according to the reference picture that following formula synthesizes to the end on the horizontal direction.
G(x,y)=q
1f(x,y)*ψ
2,0+q
2f(x,y)*ψ
3,0+q
3f(x,y)*ψ
4,0
According to the regularity of distribution of people's face portion organ, width and height that w=130 and h=150 represent reference picture are respectively established in the distribution of energy guestimate human eye area, and the coordinate of left eye and right eye is respectively e
1(x
1, y
1) and e
2(x
2, y
2), x
1=1.3w/4, y
1=1.6h/5, x
2=2.8w/4, y
2=1.6h/5, with these 2 be the center, obtain two human eye pre-service window E1 and the E2 of size for 0.32w * 0.32w, see Fig. 4.To each human eye pre-service window E, it is divided into three subclass, the cluster centre of each subclass of initialization is center
1=min (E), center
2=max (E), center
3=(center
1+ center
2)/2.Find new cluster centre center by the K-means algorithm
1', center
2', and center
3', center
1' represented subclass is set at white, other two subclass are set at black, if the piece that one group of white pixel is formed satisfies one of following condition, then this piece is changed to black.(1): the width of piece is less than the height of piece.(2): the pixel sum of piece is less than T (T=10).Calculate the quantity n of white piece, calculate the center of each white piece simultaneously, be designated as c
1(x, y), c
2(x, y) ..., c
n(x, y).This n center is obtained new center sequence c by the ordinate ordering
1' (x, y), c
2' (x, y) ..., c
n' (x, y), c here
1' (y)≤...≤c
n' (y), if n=1, then e (x, y)=c
1' (x, y).If n>=2, then e (x, y)=c
2' (x, y).
If the center of two eyes that obtain is respectively e
l(x, y) and e
r(x, y), e wherein
l(x, y) expression left eye center, e
r(x, y) expression right eye center.n
lThe number of white piece in the expression left eye pre-service window, n
rIf the number of white piece in the expression right eye pre-service window is e
l(x, y) and e
r(x, y) 2 formed horizontal sextant angles are spent greater than 30, and n
l>n
rThe time, eye center position e
l(x y) is removed, n
l<n
rThe time, eye center position e
r(x y) is removed, and recomputates the position of new eye center, and is last, and (x y) is the center, and the rectangular window that cuts 31 * 13 pixel sizes on people's face gray level image is as human eye window EW, and detailed process is seen Fig. 5 with the eye position e that obtains.
To each pixel among the eyes window EW, ask 3 * 3 neighborhoods around this pixel pixel value and, then with and value substitute the gray-scale value of this pixel, see Fig. 6.To EW scanning one time, the human eye video in window that newly obtains is designated as NI with this field operator
(3,3), at NI
(3,3)In, making uses the same method obtains 5 * 5 neighborhood images of this image, is designated as NI
(5,5)Find NI
(5,5)In have minimum gradation value the position, be designated as p
Min(x, y), then pupil center location p (x, y)=p
Min(x, y).
Claims (1)
1. the human eye accurate positioning method based on half-tone information is characterized in that this localization method comprises the detection of people's face, grey level histogram analysis, figure image intensifying, Gabor wavelet filtering, cluster analysis and neighborhood operation step; Specific as follows:
(1) people's face detects: with the Adaboost algorithm digital picture of input is carried out people's face detecting operation, obtain the position and the size of people's face;
(2) grey level histogram analysis: detected facial image unification is zoomed to 130 * 150 pixel sizes carry out the gray processing operation, the rectangular area that a size selecting the facial image middle position is 31 * 13 pixels is as the object of colour of skin sample analysis, if I is (i, j) be (i in the facial image, j) gray-scale value of locating, m is the average gray of rectangular area, if m<180, then in the grey level histogram of this rectangular area, find [m-10, m+10] scope in the maximum gray-scale value of occurrence number as the gray-scale value si of the colour of skin, facial image is carried out the grey level histogram analysis, determine the tonal range of area of skin color in people's face;
(3) figure image intensifying: the gray-scale value si of definite colour of skin, use following formula that image is strengthened operation, new gray level image (i, the gray-scale value of j) locating be designated as I ' (i, j),
c
1=2c
2,c
2=2.35
If m 〉=180 then do not strengthen operation to image, promptly I ' (i, j)=I (i, j);
(4) Gabor wavelet filtering: for the facial image after strengthening, use the Gabor small echo that image is carried out filtering operation, use formula to be
In the formula, z=(x, y) the spatial domain coordinate of remarked pixel point, || || the computing of expression norm,
k
v=k
Max/ f
v,
V ∈ 0,1 ..., 4} and μ ∈ 0,1 ..., 7} represents the direction and the yardstick of Gabor wave filter respectively, in 40 real part images that produce after Gabor filtering, selects μ=0, v={2,3,4}, promptly on the horizontal direction three of the yardstick maximum real part images according to formula:
G(x,y)=q
1f(x,y)*ψ
2,0+q
2f(x,y)*ψ
3.0+q
3f(x,y)*ψ
4.0
The reference picture that synthesizes to the end;
(5) cluster analysis: according to the regularity of distribution of people's face portion organ, width and height that w=130 and h=150 represent reference picture are respectively established in the distribution of energy guestimate human eye area, and the coordinate of left eye and right eye is respectively e
1(x
1, y
1) and e
2(x
2, y
2), x
1=1.3w/4, y
1=1.6h/5, x
2=2.8w/4, y
2=1.6h/5, with these 2 be the center, obtain two human eye pre-service window E1 and the E2 of size for 0.32w * 0.32w; To each human eye pre-service window E, it is divided into three subclass, the cluster centre of each subclass of initialization is center
1=min (E), center
2=max (E), center
3=(cener
1+ center
2)/2; Find new cluster centre center by the K-means algorithm
1', center
2', and center
3', center
1' represented subclass is set at white, other two subclass are set at black; If the piece of one group of white pixel composition satisfies one of following condition, then this piece is changed to black, and 1. the width of piece is less than the height of piece; 2. the pixel sum of piece is less than T, T=10; Calculate the quantity n of white piece, calculate the center of each white piece simultaneously, be designated as c
1(x, y), c
2(x, y) ..., c
n(x, y); This n center is obtained new center sequence c by the ordinate ordering
1' (x, y), c
2' (x, y) ..., c
n' (x, y), if n=1, then e (x, y)=c
1' (x, y); If n>=2, then e (x, y)=c
2' (x, y); If the center of two eyes that obtain is respectively e
l(x, y) and e
r(x, y), e wherein
l(x, y) expression left eye center, e
r(x, y) expression right eye center; n
lThe number of white piece in the expression left eye pre-service window, n
rIf the number of white piece in the expression right eye pre-service window is e
l(x, y) and e
r(x, y) 2 formed horizontal sextant angles are spent greater than 30, and n
l>n
rThe time, eye center position e
l(x y) is removed, n
l<n
rThe time, eye center position e
r(x y) is removed, and recomputates the position of new eye center, and is last, and (x y) is the center, and the rectangular window that cuts 31 * 13 pixel sizes on people's face gray level image is as human eye window EW with the eye position e that obtains;
(6) neighborhood operation: to each pixel among the eyes window EW, ask 3 * 3 neighborhoods around this pixel pixel value and, then with and value substitute the gray-scale value of this pixel; The human eye video in window that newly obtains is designated as NI
(3,3), at NI
(3,3)In, making uses the same method obtains 5 * 5 neighborhood images of this image, is designated as NI
(5,5)Find NI
(5,5)In have minimum gradation value the position, be designated as p
Min(x, y), then pupil center location p (x, y)=p
Min(x, y).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009100947640A CN101615292B (en) | 2009-07-24 | 2009-07-24 | Accurate positioning method for human eye on the basis of gray gradation information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009100947640A CN101615292B (en) | 2009-07-24 | 2009-07-24 | Accurate positioning method for human eye on the basis of gray gradation information |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101615292A CN101615292A (en) | 2009-12-30 |
CN101615292B true CN101615292B (en) | 2011-11-16 |
Family
ID=41494916
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2009100947640A Expired - Fee Related CN101615292B (en) | 2009-07-24 | 2009-07-24 | Accurate positioning method for human eye on the basis of gray gradation information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101615292B (en) |
Families Citing this family (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101887518B (en) * | 2010-06-17 | 2012-10-31 | 北京交通大学 | Human detecting device and method |
CN102799888B (en) * | 2011-05-27 | 2015-03-11 | 株式会社理光 | Eye detection method and eye detection equipment |
CN102880292A (en) * | 2012-09-11 | 2013-01-16 | 上海摩软通讯技术有限公司 | Mobile terminal and control method thereof |
US20140079319A1 (en) * | 2012-09-20 | 2014-03-20 | Htc Corporation | Methods for enhancing images and apparatuses using the same |
CN103632136B (en) * | 2013-11-11 | 2017-03-29 | 北京天诚盛业科技有限公司 | Human-eye positioning method and device |
CN105447822B (en) * | 2014-06-27 | 2019-07-02 | 展讯通信(上海)有限公司 | Image enchancing method, apparatus and system |
CN105320919A (en) * | 2014-07-28 | 2016-02-10 | 中兴通讯股份有限公司 | Human eye positioning method and apparatus |
CN104866587A (en) * | 2015-05-28 | 2015-08-26 | 成都艺辰德迅科技有限公司 | Data mining method based on Internet of Things |
CN105930762A (en) * | 2015-12-02 | 2016-09-07 | 中国银联股份有限公司 | Eyeball tracking method and device |
JP6744123B2 (en) * | 2016-04-26 | 2020-08-19 | 株式会社日立製作所 | Moving object tracking device and radiation irradiation system |
CN106127160A (en) * | 2016-06-28 | 2016-11-16 | 上海安威士科技股份有限公司 | A kind of human eye method for rapidly positioning for iris identification |
CN106778538A (en) * | 2016-11-28 | 2017-05-31 | 上海工程技术大学 | Intelligent driving behavior evaluation method based on analytic hierarchy process (AHP) |
CN107895157B (en) * | 2017-12-01 | 2020-10-27 | 沈海斌 | Method for accurately positioning iris center of low-resolution image |
-
2009
- 2009-07-24 CN CN2009100947640A patent/CN101615292B/en not_active Expired - Fee Related
Also Published As
Publication number | Publication date |
---|---|
CN101615292A (en) | 2009-12-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101615292B (en) | Accurate positioning method for human eye on the basis of gray gradation information | |
CN103914676B (en) | A kind of method and apparatus used in recognition of face | |
CN108491784B (en) | Single person close-up real-time identification and automatic screenshot method for large live broadcast scene | |
CN101339607B (en) | Human face recognition method and system, human face recognition model training method and system | |
CN102708361B (en) | Human face collecting method at a distance | |
CN104463877B (en) | A kind of water front method for registering based on radar image Yu electronic chart information | |
CN101930543B (en) | Method for adjusting eye image in self-photographed video | |
US20150131889A1 (en) | Necrotic cell region detection apparatus and method of the same, and non-transitory computer readable storage medium to store a necrotic cell region detection program | |
CN103116763A (en) | Vivo-face detection method based on HSV (hue, saturation, value) color space statistical characteristics | |
CN101430759A (en) | Optimized recognition pretreatment method for human face | |
CN101739546A (en) | Image cross reconstruction-based single-sample registered image face recognition method | |
CN109740572A (en) | A kind of human face in-vivo detection method based on partial color textural characteristics | |
CN106650606A (en) | Matching and processing method for face image and face image model construction system | |
CN107066952A (en) | A kind of method for detecting lane lines | |
CN102982542A (en) | Fundus image vascular segmentation method based on phase congruency | |
CN103034838A (en) | Special vehicle instrument type identification and calibration method based on image characteristics | |
CN109190456B (en) | Multi-feature fusion overlook pedestrian detection method based on aggregated channel features and gray level co-occurrence matrix | |
CN111507426A (en) | No-reference image quality grading evaluation method and device based on visual fusion characteristics | |
CN110728185B (en) | Detection method for judging existence of handheld mobile phone conversation behavior of driver | |
CN109359577A (en) | A kind of Complex Background number detection system based on machine learning | |
KR20120089527A (en) | Method and system for iris recognition | |
CN105893916A (en) | New method for detection of face pretreatment, feature extraction and dimensionality reduction description | |
CN110689003A (en) | Low-illumination imaging license plate recognition method and system, computer equipment and storage medium | |
CN115082776A (en) | Electric energy meter automatic detection system and method based on image recognition | |
CN105447457A (en) | License plate character identification method based on adaptive characteristic |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
C17 | Cessation of patent right | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20111116 Termination date: 20120724 |